What reduces hallucinations in answers?
Why B is the answer
Grounding responses with a knowledge base
Grounding responses with a knowledge base gives the model real content, which reduces hallucinations.
In detail
- Grounding gives the model real source content.
- Grounding reduces this risk, but does not remove it.
Use case
An illustrative scenario, not a real incident.
An assistant must answer from its product manuals, so it can cite them, and it invents less.
See it in different ways
flowchart LR A["User input"] --> B["AWS WAF"] B --> C["Input guardrail"] C --> D["Foundation model"] D --> E["Output guardrail"] E --> F["Grounding check"] F --> G["Response"]
mindmap
root((RAG))
Ingest
Sync data source
Detect personal data
Chunk
By structure
Hierarchical
Search
Vector
Keyword
Rank
Rerankerflowchart LR S0["Question"] --> S1["Retrieve from knowledge base"] --> S2["Grounded answer"]
Key point: Ground answers in knowledge.
Why the other options are wrong
- A. Increasing the output length, so that answers are more complete
- Longer output does not add facts.
- C. Raising the temperature, so that the model considers more options
- Higher temperature can make it worse.
- D. Removing the source documents from the retrieval pipeline
- That removes the grounding.